◀ Siri 2:15 PM Tue 25 Aug Note Title: The Evolution of AI 25 August 2026 at 2:15 PM

What is Artificial Intellegence?

AI is a science of making machins perform tasks that normally require human intellegence.

1950 – Can machine think?

Alan Turing → The turing test

1955 – John McCarthy (Artificial Intellegence)

Researchers came together with an ambitios belief →

“ Every aspect of learning and intellegence could, in principle, be described precisely enough for a machine to simulate. ”

★ After 1955 AI winter were there for a long period.

1986 → Synthetic Intellegence

1997 → Deep blue defeated Garry Kasparov.

IBM developed

Best chess player that time

Have machines become smarter than human?

# Rule based AI (1950 – 1980)

“ Intellegence is simply a collection of Rules ”

Ex – Spam Detector: Hardcoded checks for keywords like “Free”, “$$$”, “LOTTERY”.

→ “Free”, “$$$”, “LOTTERY”

Medical Expert System:

                → If (Fever && Cold && Body Ache) → FLU
              

“ Expert systems were built by human experts writing a lot of rules. ”

# Machine Learning

❌ Rules ✔ Examples

→ Cat vs Dog: Train using 1 million labelled pictures.

→ Spam Detection: Adapts to clever variations: F-R-E-E , FREEE , FREE!!!

Model learns

Training data

Predictions

Learn patterns

# Deep Learning

– Neural Networks → Important breakthrough for AI

Can computer learn features themselves?

Instead of human telling “this is an eye” — can computer discover “eye” automatically?

Image Recognition

Speech Recognition

Translation

GPU revolution + Internet + Large dataset

⇒ Difference between Machine Learning & Deep Learning: Deep Learning automatically discovers hierarchical feature representations from raw data without manual feature engineering!

# Computer Vision Revolution

– The dataset: ‘ImageNet’

– Trained large, large neural networks on powerful GPUs.

AlexNet (2012)

[ Alex Krizhevsky & team

[ Deep Convolutional Neural Network (CNN)

[ It was a big, big leap — reduced ImageNet error from 26% to 15%!

> Face Unlock

> Self Driving cars

> X Rays & Diagnostics

> Shopping app products

“ Machine could actually see ”

# Natural Language Processing (NLP)

“ I saw a man with a telescope ”

→ (Did I use the telescope, or was the man holding it?)

“ The chicken is ready to eat ”

→ (Is the chicken eating food, or is it cooked and ready to be eaten?)

0. Bag of Words: Simple word frequencies; loses word order and grammatical context.

1. N-grams & Statistical Algorithms: Local sequential patterns (pairs/triplets of words).

2. RNN (Recurrent Neural Networks): Sequential token processing; struggled with long-term memory & vanishing gradients.

3. LSTM (Long - Short Term Memory): Gated memory cells → much, much better RNN!

↳ much much better RNN (solving sequential bottlenecks with Self-Attention)

# Transformers (2017) {Big revolution in AI}

→ Most remarkable history of mankind.

“ Attention is all you need ” → Groundbreaking Research Paper by Google

“ The lion did not cross the river because it cannot swim ”

⬆ Self-Attention dynamically binds the pronoun ‘it’ to ‘lion’ rather than ‘river’ based on contextual probability!

# Large Language Models (LLMs)

“ Transformers trained on very very very very large large large data ”

# Generative AI

Old AI Paradigm (Discriminative)

• Classification

• Predictions

• Recommendations

Modern AI Paradigm (Generative)

• Generate (Multi-Modal)

→ Text, Images, Videos, Audio, Code & Documents!

# The ChatGPT Moment (NOV 2022)

– Conversation: Intuitive natural language interface accessible to anyone.

– Made it public: Fastest adopted consumer application in history (100M+ users).

– RLHF: Reinforcement Learning from Human Feedback for alignment, safety & helpfulness.

– Memory & Context: Long-horizon reasoning across millions of tokens.

– Led the whole AI movement: Foundation models powering everyday intelligence.

# Multimodal AI

🖼️ Images 🎙️ Voice & Audio 🎥 Videos 💬 Languages & Code 📄 Documents

“ Almost every form of data ”

# The AI today

Earlier AI could only answer static questions. Today, modern AI agents can:

⚡

Think & Reason

Chain of Thought & planning

⚡

Work autonomously

Multi-step agent loops

⚡

Plan

Decompose complex goals

⚡

Make decisions

Self-correct & adapt

⚡

Call APIs

Integrate external services

⚡

Complete tasks

End-to-end execution

⚡

Write code

Full-stack software engineering

⚡

Has context

Understand deep nuance

⚡

Understand

Multimodal cross-reasoning

⚡

Solve problems

Math, science & logic

⚡

Remember

Persistent & episodic memory

⚡

Build and Test

Automated CI/CD & verification

⚡

Search web

Real-time information retrieval

⚡

Use tools

Terminal, IDE, browser & files

1950 – Alan Turing 1956 – John McCarthy 2012 – Deep Learning 2017 – Transformers 2022 – ChatGPT 2026+ – Agentic AI

Analysis: Are Machines Smarter Than Humans?

The answer depends on whether we measure Narrow Specialization vs Artificial General Intelligence (AGI).

Dimension	Human Brain 🧠	AI Systems 🤖
Processing Speed	~100 Hz biological signal transmission	TeraFLOPS & PetaFLOPS parallel compute
Memory Capacity	Selective, associative & reconstructive	Petabytes of lossless retrieval & 2M+ token contexts
Specialized Mastery (Chess/Math/Code)	Human champions surpassed	Superhuman performance (Deep Blue, AlphaGo, o1/o3)
Autonomous Tool & API Usage	Manual manipulation	Instantaneous programmatic agent orchestration
Generalization & Empathy	Unmatched biological consciousness, emotional depth & moral judgment	Probabilistic synthesis, synthetic empathy, hallucination risk

“ Scaling laws and agent architectures prove that machines excel at computational scale, synthesis, and rapid tool orchestration — while humans remain the architects of intent, ethics, and profound creative insight. ”

Conclusion: The future is Collaborative Superintelligence — Humans + AI Agents working together!

Table of Contents (AI Master Notes)